The Artificial Intelligence Show – Episode #199 AI Answers: Do Custom GPTs Still Matter? AI Output Validation, 2026 Job Disruption, Preventing Burnout & Build vs. Buy
Release Date: February 26, 2026
Hosts: Paul Roetzer (CEO, Marketing AI Institute & SmartRx), Kathy McPhillips (CMO, SmartRx)
Episode Overview
This episode of The Artificial Intelligence Show is part of the special "AI Answers" series, where Paul and Kathy address real questions from their monthly live Intro to AI and Scaling AI classes. The conversation focuses on what business leaders and professionals are struggling with right now in AI—from validation of AI outputs and the real value of custom GPTs, to workplace disruption, employee burnout, and making smart decisions about building versus buying new AI solutions.
The hosts aim to bring both practical Q&A and future-forward perspectives, with a clear-eyed look at the accelerating pace of AI's impact on enterprises and careers.
Key Discussion Points and Insights
The Purpose & Structure of AI Answers (03:00–06:50)
- AI Answers tackles dozens of audience questions that don’t make it into the live Q&A at their virtual events.
- Questions are synthesized, deduplicated, and ordered for flow; Paul answers them "raw," without seeing them beforehand.
- The goal: Accelerate AI literacy and provide actionable, up-to-the-minute guidance for professionals at all stages.
1. Structured Prompting in AI Platforms (07:11)
- Q: Do you need to use structured prompting every time, or does the AI remember?
- Paul: Prompting is an ongoing experiment; structured prompts are useful but experimenting with both detailed and simple prompts can yield surprising—and sometimes better—results.
- "Oftentimes you get surprised. So I’ll often go to a project and say, ‘Here’s what I’m trying to do. What do you think I should include in it?’" (08:35)
- Prompt libraries are still valuable for reference and overcoming "blank page" syndrome; save your best prompts for future use, especially across different model generations. (09:50)
2. Do Custom GPTs Still Matter? (10:57)
- Q: If ChatGPT can answer directly, what’s the value in custom GPTs?
- Paul: Custom GPTs excel at consistency for repeated tasks and allow for shared use within teams.
- Example: Paul’s AI learning assistant "Aila" was trained on his organization's design principles, streamlining course development.
- "Every time I needed something, I didn’t have to go back in and say, ‘Here’s what AI Academy is, here’s what I’m trying to do.’" (11:38)
- Shared GPTs boost team efficiency by giving everyone access to the same preloaded knowledge and workflows (Kathy, 14:05).
3. SaaS and Model-Agnostic Approaches (14:36)
- Q: Are SaaS providers allowing "bring your own model"?
- Paul: Most major SaaS companies don’t build their own models—they license access from AI labs like OpenAI, Google, Anthropic, etc., often using whichever model is most cost-effective or best for the task.
- "Software companies build these models into their software... If you go into HubSpot and you’re interacting with an AI... that is not HubSpot’s model." (15:00)
- Expect increasing model-agnosticism as a hedge against cost and reliability.
4. Model Updates Impact Output Quality (17:23)
- Q: Are newer models less consistent in tone/voice for custom GPTs?
- Yes—changes in model architecture (e.g., GPT 5.2 vs. 4) can cause noticeable shifts in tone, style, or capability.
- "This is a real problem... You need to go in and experiment with them and make sure they’re still doing what they’re supposed to do." (18:00)
- Users must monitor and adjust system instructions with every major update. There’s no way to prevent these shifts; model creators dictate behaviors with each release.
5. Validating AI Output (20:50)
- Q: What’s the best process for validating AI-generated output?
- The only real safeguard is strong human oversight.
- "There is no shortcut and nor do I think there should be a shortcut for verification. I think that we should slow down and publish good quality stuff that we have verified and that humans have signed off on." (20:51)
- It’s fine to use a secondary LLM for first-pass validation, but humans must critically evaluate sources and outputs before publishing.
6. Tools for Building AI Agents (23:29)
- Q: How can someone start building practical AI agents?
- Start with your core stack—Microsoft Copilot, Google AI Studio, Salesforce, etc.—which increasingly enable rules-based "agent" creation.
- General-purpose coding agents like Claude Code, Lovable, Agent AI, and Gemini are rapidly democratizing development.
- True autonomy in non-coding knowledge work is still rare but advancing quickly; watch this space into 2026.
7. Job Disruption for Knowledge Workers (26:09)
- Q: Will knowledge work get disrupted like SaaS/development?
- The tech is ready, but "human friction" and enterprise inertia mean systemic displacement will take longer than the tech curve suggests.
- "If that was true today, it would take five years for these Enterprises to do something about it." (27:30)
- Job loss is accelerating, especially into 2026, but most companies are still unprepared.
8. Preventing Burnout in AI Transitions (30:03)
- Q: As AI increases productivity, how do leaders prevent burnout?
- Early AI adopters (often a minority on teams) may overproduce and feel pressure; leaders must recognize value-creation over mere output, offer time-off or grace, and explicitly manage workloads.
- "Sometimes you have to look at the value of the output, not the time you spent on it, and you have to be able to give yourself time back." (34:33)
- Intrinsically motivated team members are susceptible to burnout unless supported.
9. Roles and Skills Most at Risk (36:31)
- Q: Which jobs are most likely to become obsolete, and how can leaders help?
- Emotional intelligence, critical thinking, asking the right questions, and strong writing remain “safe-ish.”
- Vulnerability isn’t about titles—it’s about which labor markets are targeted by VC and software; what can be automated, will be, if there’s enough economic incentive.
- "There’s no role that’s actually safe... The byproduct of that effort is that fewer humans are needed to do the same work." (39:45)
10. Analytics: BI vs. AI-First (42:03)
- Q: Should leaders keep investing in traditional BI or shift to AI-first?
- Ideal is conversational, insight-surfacing AI layered atop existing platforms.
- Waiting for vendors to unlock native intelligence versus connecting to LLMs like Claude or Gemini directly is a tough architectural choice.
- "All I have is data. I’m thinking about... What are the questions I want answered... then what is the best way to architect that?" (43:12)
11. Build vs. Buy for AI Capabilities (45:34)
- Q: What should companies build vs. buy in the AI era?
- Old rule: build what’s core, buy what’s ancillary. In AI, custom model development is rarely viable for all but the biggest players.
- Today, find partners with strong roadmaps; tailor and customize, but don’t expect to out-build the top labs.
- "We found a vendor with an AI roadmap that we believed in... I could have stopped and spent a million dollars and tried to build our own learning management system, but..." (47:41)
12. Competitive Advantage for Independent Agencies (49:11)
- Q: How do AI-forward agencies maintain an edge?
- With consultancies and big tech investing billions, the only lasting advantage is to stay on the leading edge—constant reassessment, reinvention, and focus on agility and impact.
- "I think there’s going to be these... S curves of competitive advantage. It’s going to be like, ‘Oh, you’re the leader in X’... and then it just like plummets..." (51:32)
13. Spotting AI-Generated Work (52:41)
- Q: How can managers (and teams) spot when work has been AI-generated without deep engagement?
- Paul: Ask for walkthroughs, drill with follow-up questions, test whether someone has critical understanding or is passing off AI slop.
- "If you’re handing something in or publishing something publicly, you better have the confidence... that if put on a stage... you could actually confidently answer those questions." (54:18)
14. Ads Entering AI Platforms (55:01)
- Q: What impact will ads in AI assistants have?
- Minimal for business users—paid/enterprise tiers will largely remain ad-free for now. Brands may want to be ready if AI assistants prove a viable ad channel.
- "I don’t think largely they’re going to change much of anything for business users. I think they’ll largely be excluded... for a while." (55:11)
15. The One AI Superpower for Business Leaders (56:48)
- Q: If you could give every leader one AI-powered superpower?
- Deep situational awareness—genuine, nuanced understanding of AI’s current state and trajectory.
- "Way too many CEOs have no real concept of what’s happening... If the leaders truly understood the moment... they would be moving way faster..." (56:48)
- With this, everything else (strategy, urgency, resource allocation) would follow.
Notable Quotes & Memorable Moments
-
On validation:
"These things make mistakes, they hallucinate, they have errors in them. And just because we can take shortcuts on the creation of it doesn’t mean we can take shortcuts on the verification..." (20:51) – Paul -
On model updates:
"You’re going to have to constantly deal with this... when new models come out and they get infused into the software you use or the GPTs or gems you’ve built." (19:40) – Paul -
On enterprise inertia:
"We could have AGI today, we could agree as a society that we have reached artificial general intelligence. And I don’t know that it would change anything within enterprises." (28:12) – Paul -
On agency competitive advantage:
"I don’t know there’s any service I could guide an agency to offer today that... is going to be obsoleted in 18 months... It’s... S curves of competitive advantage." (51:16) – Paul -
Managing burnout:
"Sometimes you have to look at the value of the output, not the time you spent on it, and you have to be able to give yourself time back." (34:33) – Paul
Timestamps of Key Sections
- 07:11 – Structured Prompting and Prompt Libraries
- 10:57 – Value of Custom GPTs
- 14:36 – SaaS Providers and Model Agnosticism
- 17:23 – Model Updates Impacting Voice and Tone
- 20:50 – Validating AI Output
- 23:29 – How to Build AI Agents
- 26:09 – Job Disruption for Knowledge Workers
- 30:03 – Preventing AI Burnout
- 36:31 – Jobs and Skills Most at Risk
- 42:03 – BI vs. AI-First Analytics
- 45:34 – Build vs. Buy Decision Framework
- 49:11 – Lasting Advantages for Agencies
- 52:41 – Spotting AI-generated Work
- 55:01 – Ads Entering AI Assistants
- 56:48 – The One Superpower All Leaders Need
Closing Sentiment
Paul and Kathy stress that fast, continuous learning and strategic adaptability are the only constants in the AI revolution—especially as roles, tools, and best practices shape-shift faster and faster. Leaders must cultivate deep awareness, experiment relentlessly, and build cultures that reward curiosity, integrity, and thoughtful risk-taking.
For more:
Visit SmarterX AI for on-demand courses, up-to-date resources, and details on live events—plus the ongoing State of AI for Business survey.
